google-cloud-go — Engineering Performance
29 engineers all time · Jan 2025 – Nov 2026 · built 2026-11-22 · GitHub
Performance snapshot
Today's rolling 90-day reading for google-cloud-go, compared with the start of the series. Pick a window to move that comparison point.
Avg. perf / dev / mo
+110.3%
0.48 → 1.01 ETV
Active engineers
−14.3%
14.0 → 12.0
Features
−16.4pp
43.3% → 26.9%
vs. Google
0.27x
0.51x → 0.27x · −73% below
google-cloud-go vs. Google
Per-engineer ETV for google-cloud-go against Google as a whole. Both lines are 90-day rolling averages scaled to a 30-day month, so they share one axis and can be read against each other at any point. Pick a window to zoom the chart to it.
Performance Composition
Each month's output split by type of work: Features (new value), Maintenance (sustaining systems), Tests, Docs, and Fixes (rework). The yellow line is output per engineer, so when it rises each engineer is delivering more, whatever the team size did. Unit: Engineering Throughput Value (ETV).
Engineering capacity
Effective engineers behind google-cloud-go, against its pre-AI baseline. Each subject has its own: google-cloud-go's is 0.48 ETV / dev / mo, its first reading in Q1 2025. Per-engineer ETV divided by that gives a capacity multiple, and that multiple applied to the engineers active in the trailing 90 days turns it into engineer-equivalents. The line is the real headcount, so the gap between line and area is what the leverage is worth. Because each baseline is its own, every subject opens at 1.0x on its first day: multiples measure improvement and are not comparable between subjects.
Knowledge concentration
How dependent is this repo on a small number of engineers? Higher top-1 share = higher key-person risk.
Baha Aiman owns 14.4 % of commits.
Behind the numbers
Written summary of the work completed each month.
No monthly reports available yet.
Top engineers
Most impactful commits
Top 10 by ETV in the all-time window.
- 6.5ETVfeat(spanner): complete location-aware routing resilience and observability (#14418) ## Summary This change expands the location-aware routing path in the Spanner Go client with endpoint lifecycle management, stronger endpoint health handling, retry-aware exclusion, latency-aware replica selection, and route-selection tracing. ## What Changed - enabled location-aware routing automatically for experimental-host configurations while preserving env-var override behavior - added endpoint cache support for GetIfPresent, Evict, and DefaultChannel - switched endpoint health evaluation to real gRPC connectivity state, including transient-failure detection - added an endpoint lifecycle manager with: - background probing - idle eviction - endpoint recreation requests - recent transient-failure eviction tracking - integrated lifecycle handling into routing and affinity fallback paths - added request-id keyed one-shot endpoint exclusion for RESOURCE_EXHAUSTED retries - expanded skipped_tablet_uid reporting for transient failures and recent transient-failure evictions, with deduplication - added OpenTelemetry route-selection span attributes and events - updated focused tests and goldens for the new routing behavior ## Behavioral Notes - stopped or transient-failure endpoints are now treated differently from simply non-ready endpoints during tablet selection - retry attempts can avoid previously selected routed endpoints after RESOURCE_EXHAUSTED - leader-preferred requests still favor close leaders, but can fall back to a closer non-leader replica when the leader is too far away - route selection is now observable through tracing attributes and eventsrahul2393 · 77aa4df8 · 2026-04-14
- 6.0ETVfeat(workloadmanager): add new clients (#13849) PiperOrigin-RevId: 869327994shollyman · 13288723 · 2026-02-14
- 4.5ETVfeat(spanner): move DirectPath fallback into the library with paired channel slots (#20651) Before this change, DirectPath fallback came from grpc-gcp-go's GCPFallback wrapped around a static channel pool. That had three problems: - It dialed a second CloudPath pool of NumChannels at startup, whether or not it was ever needed. - It round-robined every RPC across the active pool, so the statements of one transaction could land on different connections, and request ids reported channel 0. - The dynamic channel pool had no DirectPath fallback at all. The fixed and dynamic channel pools now handle fallback themselves: - Each pool entry owns a channel slot: its DirectPath connection plus a CloudPath connection that is dialed only on first use after the pool falls back. Sessions and transactions stay on their entry across a switch, and request ids keep the entry's real channel id. - Fallback is pool-wide and one-way. Requests already in flight finish on the path they started on, and late DirectPath results from before the switch are ignored. - The switch rule, error codes and eef metrics are unchanged. - Calls the caller cancelled and channel priming no longer count toward the error rate. A stream that delivered data before its reader stopped it counts as a success. - A client that supplies its own connection or pool has no separate CloudPath and keeps using what it was given. Not in this change: - The legacy static pool (GOOGLE_SPANNER_ENABLE_LEGACY_STATIC_CHANNEL_POOL=true) keeps GCPFallback. - The fixed pool has no fallback with the emulator. - Recovery back to DirectPath. The slot and fallback state are shaped so per-slot recovery can be added later.rahul2393 · 84b31d12 · 2026-11-20
- 4.4ETVfeat(auditmanager): add new clients (#13842) PiperOrigin-RevId: 861758611Alex Hong · 588fe0c8 · 2026-02-12
- 3.9ETVfeat(hypercomputecluster): add new client (#13423) PiperOrigin-RevId: 823382075Alex Hong · 945efa94 · 2025-12-02
- 3.8ETVfeat(iam/admin): regenerate client (#11570) This client has some handwritten code in it that causes failures when generating it normally. I generated the client and hand patched around handwritten code as to not create a breaking change. Changes made: Keep old variants of RPCs that did not generate an iterator helper to avoid breaking changes: ListRoles, QueryGrantableRoles, and QueryTestablePermissions Add variants of above RPCs that do return an iterator: ListRolesIter, QueryGrantableRolesIter, and QueryTestablePermissionsIter Preserve API for GetIamPolicy and SetIamPolicy. Fixes: #8219Cody Oss · eab87d73 · 2025-02-07
- 3.8ETVchore: onboard a new mapmanagement library (#14467)Noah Dietz · 26125607 · 2026-04-22
- 3.7ETVfeat(storage): merge support for bi-directional multiple range reads (#11377) * feat(storage): merge support for bi-directional multiple range reads * add license header * get object metadata for regular grpc reads * docs change * fix offset and remaining bytes calculation * fix(storage): use mutex for accessing concurrent vars * remove TestOpenAppendableWriterUnsupportedEmulated * Add back retry conformance tests to json * retry tests assume writes are conditionally idempoten --------- Co-authored-by: Daniel B <danielduhh@gmail.com>Brenna N Epp · b4d86a52 · 2025-01-08
- 3.7ETVchore: onboard new library appoptimize (#14393) Onboard new API client for `google/cloud/appoptimize/v1beta` Internal bug b/500129743Noah Dietz · e247d48b · 2026-04-08
- 3.5ETVrefactor(spanner): move dynamic channel pool core to internal/transport (#20919) ### Summary This PR moves the generic Dynamic Channel Pool (DCP) core implementation from [`spanner/dynamic_channel_pool.go`](spanner/dynamic_channel_pool.go) into [`spanner/internal/transport/dcp.go`](spanner/internal/transport/dcp.go) with **zero public API changes** in `cloud.google.com/go/spanner` ### Testing - Added unit tests in [`spanner/internal/transport/dcp_test.go`](spanner/internal/transport/dcp_test.go). - All existing Dynamic Channel Pool tests in [`spanner/dynamic_channel_pool_test.go`](spanner/dynamic_channel_pool_test.go) passing.Subham Sinha · 5fbc4e72 · 2026-11-18